Research On Measures Of Economic Uncertainty Based On E-Commerce Big Data

For economic individuals,uncertainty measures the unpredictable part of future.And economic uncertainty is a variable that can not be directly observed,it is often necessary to use the appropriate proxy variables to reflect the true level of the measures.Different from the traditional sampling modeling and forecasting,the method of constructing economic uncertainty measures based on big data are more oriented to the total,real-time data and make the current forecasting.Also,it has the advantage that traditional sampling data and forecasting model can’t compare with.As e-commerce data can more fully reflect the consumer demand,especially for nonphysical goods retail sales which most indicators can not fully reflect.Therefore,this thesis chooses to use e-commerce data which are combined with the relationship between the variables,and use the Graph Auto-Encoders with the principle of graph convolution to construct a measure index that can measure the economic uncertainty of China.And it is compared with the measure index of China’s economic uncertainty which is often used in the present research.In the study of corporate investment and the fluctuation of China’s stock market,respectively,experiments were carried out to replace the original measure of economic uncertainty,to observe the measurement ability of the indicators constructed in this paper for China’s economic uncertainty and the interpretability of macroeconomic phenomena.Using the data of ecommerce industry,which has a great influence on the real economy of our country,as the explanatory variable,we mine the valuable information.The selected e-commerce data is defined as the data with graph structure in non-euclidean space,and the features of the graph are extracted by Graph Auto-Encoders with unsupervised learning form.The feature of lower dimension is taken as the measure index of economic uncertainty in China.Through the existing research on the investment of Chinese enterprises and the fluctuation of the stock market,this thesis carries on the replacement index and compares the experimental results.It can be found that the indicators constructed in this thesis that can reflect China’s economic uncertainty measure have certain applicability of economic research and the ability to respond to macroeconomic phenomena finally,and the sensitivity to macroeconomic phenomena is higher than the traditional index of economic uncertainty measures.And it can verify the existence of e-commerce data valuable information can be mined.At the same time,it is proved that the Graph Auto-Encoders has excellent capability of feature extraction and representation for economic and financial variables with graph structure.